The increase in the number of processor cores has opened up new opportunities to take advantage of the parallelism available in modern data-centric applications and workloads. On the other hand, parallelising programs is still an error-prone and challenging task for the typical programmer.

Novel programming/execution models, which ease the job of the programmer while expose the hardware resources to the compiler/runtime/OS stack, are key to unleash the parallelism potential of future applications. In recent years task-based programming models (e.g. OpenMP and OpenSs) have shown to be a scalable and flexible approach to extract task parallelism from applications.